pkg/R/constructionModelesLassoMLE.R

   1 #' constructionModelesLassoMLE
   2 #'
   3 #' Construct a collection of models with the Lasso-MLE procedure.
   4 #'
   5 #'
   6 #' @param ...
   7 #'
   8 #' @return ...
   9 #'
  10 #' export
  11 constructionModelesLassoMLE = function(phiInit, rhoInit, piInit, gamInit, mini, maxi,
  12     gamma, X, Y, thresh, tau, S, ncores=3, fast=TRUE, verbose=FALSE)
  13 {
  14     if (ncores > 1)
  15     {
  16         cl = parallel::makeCluster(ncores, outfile='')
  17         parallel::clusterExport( cl, envir=environment(),
  18             varlist=c("phiInit","rhoInit","gamInit","mini","maxi","gamma","X","Y","thresh",
  19             "tau","S","ncores","verbose") )
  20     }
  21
  22     # Individual model computation
  23     computeAtLambda <- function(lambda)
  24     {
  25         if (ncores > 1)
  26             require("valse") #nodes start with an empty environment
  27
  28         if (verbose)
  29             print(paste("Computations for lambda=",lambda))
  30
  31         n = dim(X)[1]
  32         p = dim(phiInit)[1]
  33         m = dim(phiInit)[2]
  34         k = dim(phiInit)[3]
  35         sel.lambda = S[[lambda]]$selected
  36 #       col.sel = which(colSums(sel.lambda)!=0) #if boolean matrix
  37         col.sel <- which( sapply(sel.lambda,length) > 0 ) #if list of selected vars
  38         if (length(col.sel) == 0)
  39             return (NULL)
  40
  41         # lambda == 0 because we compute the EMV: no penalization here
  42         res = EMGLLF(phiInit[col.sel,,],rhoInit,piInit,gamInit,mini,maxi,gamma,0,
  43             X[,col.sel], Y, tau, fast)
  44
  45         # Eval dimension from the result + selected
  46         phiLambda2 = res$phi
  47         rhoLambda = res$rho
  48         piLambda = res$pi
  49         phiLambda = array(0, dim = c(p,m,k))
  50         for (j in seq_along(col.sel))
  51             phiLambda[col.sel[j],sel.lambda[[j]],] = phiLambda2[j,sel.lambda[[j]],]
  52         dimension = length(unlist(sel.lambda))
  53
  54         # Computation of the loglikelihood
  55         densite = vector("double",n)
  56         for (r in 1:k)
  57         {
  58           if (length(col.sel)==1){
  59             delta = (Y%*%rhoLambda[,,r] - (X[, col.sel]%*%t(phiLambda[col.sel,,r])))
  60           } else delta = (Y%*%rhoLambda[,,r] - (X[, col.sel]%*%phiLambda[col.sel,,r]))
  61             densite = densite + piLambda[r] *
  62                 det(rhoLambda[,,r])/(sqrt(2*base::pi))^m * exp(-diag(tcrossprod(delta))/2.0)
  63         }
  64         llhLambda = c( sum(log(densite)), (dimension+m+1)*k-1 )
  65         list("phi"= phiLambda, "rho"= rhoLambda, "pi"= piLambda, "llh" = llhLambda)
  66     }
  67
  68     # For each lambda, computation of the parameters
  69     out =
  70         if (ncores > 1)
  71             parLapply(cl, 1:length(S), computeAtLambda)
  72         else
  73             lapply(1:length(S), computeAtLambda)
  74
  75     if (ncores > 1)
  76         parallel::stopCluster(cl)
  77
  78     out
  79 }